AI Regulation Tracker  /  Copyright policy

Australia rules out an AI text-and-data-mining copyright exception, leaving training a licensing question

The Australian Government confirmed it will not create a text-and-data-mining exception to copyright, so training AI on Australian books, music, journalism, images, or code requires a licence or carries infringement risk.

By Anthony Guerriero  ·  Reviewed by The Leveraged Years Editorial Desk  ·  Published July 7, 2026  ·  Last updated July 7, 2026

Australia has settled one of the most contested questions in AI copyright policy, and it did so against the interests of model developers. On October 26, 2025, the Attorney-General confirmed the Government will not introduce a text-and-data-mining exception to the Copyright Act 1968. A text-and-data-mining exception would have let developers copy and process copyright works to train AI systems without permission or payment. That option is now off the table. For anyone training models on Australian material, the practical rule is simple. Use requires a licence, or it carries infringement risk.

What the Government decided

The decision follows a consultation that drew more than 400 submissions, with creative-industry bodies opposing any free-use carve-out. Rather than legislate an exception, the Attorney-General's Department is continuing its work through the Copyright and Artificial Intelligence Reference Group, known as CAIRG, a body it convened to examine copyright and AI. The Government framed the choice as providing certainty to Australian creators, and it kept the existing copyright framework intact. There is no new statutory permission for AI training, and none is proposed.

The reform debate had asked whether Australia should follow jurisdictions that allow copying for text and data mining, which supporters argued would let developers train competitive models on local data. Peak creative bodies, including music and author organisations, argued the opposite, warning that a broad exception would let developers use their members' work for free and without consent. The Government sided with rights holders on the specific question of a general exception, while leaving licensing terms to be worked out. That leaves a policy gap the CAIRG process is meant to fill, rather than a settled market.

What CAIRG is working on next

The department has set out continuing work in three areas. The first is licensing, meaning fair and legal avenues for using copyright material in AI training, which CAIRG is examining through further consultation with its members. The second is certainty about how copyright law applies to material generated using AI, an unresolved question for outputs as well as inputs. The third is enforcement, including the possible establishment of a small-claims copyright forum to give rights holders a cheaper route to enforce existing rights. These consultations continue through 2026. Separately, the Productivity Commission has been examining AI and the digital economy, with a final report expected in early 2026, so developers and rights holders should watch both tracks.

What it does not do

The announcement does not create a licence for anyone, and it does not settle pending questions about how existing law applies to past training. It confirms a policy direction and rules out one specific reform. It does not, on its own, change liability standards, set royalty rates, or create the small-claims forum, which remains under consideration rather than in force. It also does not resolve whether and how copyright subsists in AI-generated output. Firms should read the decision as removing an anticipated defence, not as delivering a finished licensing regime.

The compliance exposure for AI developers

For model trainers, the exposure is concrete. Copying Australian books, music recordings, news articles, photographs, artwork, or software into a training corpus without authorisation engages the reproduction rights in the Copyright Act 1968, and there is no dedicated exception to rely on. That places the burden on developers to obtain licences, to filter Australian works out of their datasets, or to accept the risk of an infringement claim. Content owners and publishers gain leverage, because the Government has effectively confirmed that their permission is required. Licensing teams on both sides should expect negotiation over training rights to become a standard part of dataset procurement.

Why US firms should track this

For a US reader, the direct effect is that a US developer training on Australian copyright material cannot invoke an Australian text-and-data-mining exception, because none exists. The reach is also indirect. Australia has now aligned with a licensing-first stance closer to the position taken in parts of the creative sector globally, and its rejection of a broad exception adds to the international signal that training on protected works is a licensing question, not a free-use one. US firms operating in or sourcing content from Australia should assume they need permission, and should track the CAIRG licensing consultations and the Productivity Commission report as indicators of where the rules settle.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.